The AI Page That Should Have Just Said No

A button that looked ready and wasn't

I hit one of my own admin pages recently -- the one that's supposed to generate images with AI -- and asked myself a simple question before clicking anything: is there even an AI provider connected that can actually do this? The page gave no way to tell. It looked exactly as functional as it would if everything were working. It wasn't.

The gap between "healthy" and "can do the thing"

The system was checking whether some AI provider was reachable, and reporting "healthy" as long as one was -- but the one that was reachable only handled text, and the page needed images. A system can be perfectly healthy at the one thing it does and still be completely unable to do the thing you're about to ask it for, and a health check that doesn't ask "healthy at what, specifically" will lie to you in exactly that gap.

What a page like that should do instead

The fix wasn't to make the AI more reliable -- it was to make the page honest before the click, not after. If the capability isn't there, say so up front, point at what would fix it, and hand over a manual path so the work isn't just dead in the water. A blocked feature should look blocked. It should never look ready.

That's a broader lesson than one admin page: any tool you're paying for that "uses AI" owes you the same honesty. If it can't do the specific thing you need right now, it should tell you before you've spent the time finding out the hard way.